isa-95 friendly rtn models for industrial production...

Download ISA-95 Friendly RTN Models for Industrial Production egon.cheme.cmu.edu/ewo/docs/ABB_Castro_CMU_EWO_Mar_2016.pdfISA-95 Friendly RTN Models for Industrial Production Scheduling Pedro

Post on 06-Mar-2018

219 views

Category:

Documents

5 download

Embed Size (px)

TRANSCRIPT

  • ISA-95 Friendly RTN Models for Industrial Production Scheduling

    Pedro M. CastroIgnacio E. GrossmannIiro Harjunkoski

  • Motivation EWO aims to simultaneously account

    for KPI across multiple business units Integration of supply chain management,

    production control, planning & scheduling

    Need to efficiently transfer data and information between different systems Focus on production management

    system and scheduling solution

    ISA-95 standard can act asdata-exchange platform Goal: Adopt standard in a way

    that fulfills requirements of mostcommon scheduling problems

    March 9, 2016 2Enterprise-Wide Optimization Meeting: ABB Project Overview

  • Generic scheduling approach Resource-Task Network

    Process representation Production recipe & network topology

    Mathematical formulations Discrete- & continuous-time

    Key success drivers Modeling paradigm easily understood by business stakeholders Flexible approach, easily modified when new information becomes available

    Challenge Develop RTN models complying to the ISA-95 standard

    March 9, 2016 3Enterprise-Wide Optimization Meeting: ABB Project Overview

    A_Mix_160 min

    A_Mix_270 min

    RM1

    RM2

    A_React120 min

    A_Pack30 min

    S1A S2A A

    Waste

    Mix1

    Mix2

    ReactorPackLine

    Power

    200 kg

    200 kg

    20 kg320 kg

    1200 kW

    1000 kW

    50 kW 10 kW

  • ISA-95 data implementation Variety of complex XML files

    Process segment information Equipment Material Personnel Operations capability Operations definition Operations schedule

    March 9, 2016 4Enterprise-Wide Optimization Meeting: ABB Project Overview

    Mixing Reaction Packaging

  • Set Description Elements B2MML file (XML implementation ISA-95)P1(SG,SG) Immediate precedence Mixing.Reaction, Reaction.Packaging ProcessSegmentInformationP2(SG,ID) Segment belongs to class Mixing.Mixer, Reaction.Reactor, Packaging.Pack ProcessSegmentInformation

    P3(ID,EQ) ID belonging to equipment Mixer.Mixer1, Mixer.Mixer2, Reactor.Reactor1, Pack.Packing1, Pack.Packing2 EquipmentInformation

    Mapping data with RTN process model Sets definition

    Parameters from production recipe

    March 9, 2016 5Enterprise-Wide Optimization Meeting: ABB Project Overview

    Set Description Elements B2MML file (XML implementation ISA-95)MT Materials Power, RM1, RM2, Waste, A, B, C MaterialInformationEQ Equipment Mixer1, Mixer2, Reactor1, Packing1, Packing2, Packing3 EquipmentInformationID Class ID Mixer, Reactor, Pack EquipmentInformationOP Operations ProdA, ProdB, ProdC OperationsDefinitionInformationSG Segments Mixing, Reaction, Packaging ProcessSegmentInformation

    MatReq1(OP,SG,MT) RM1 RM2 Waste A B CProdA.Mixing -200 -200

    ProdA.Reaction 20ProdA.Packaging 320

    ProdB.Mixing -500ProdB.Reaction 50

    ProdB.Packaging 400ProdC.Mixing -150 -300

    ProdC.Reaction 15ProdC.Packaging 380

    OperationsDefinitionInformation.xml

    MatReq2(OP,EQ,MT)

    Mixer1.Power

    Mixer2.Power

    Reactor1.Power

    Packing1.Power

    Packing2.Power

    Packing3.Power

    ProdA -1200 -1000 -50 -20 -10ProdB -900 -120 -12 -12 -10ProdC -800 -750 -50 -10 -10 -10

    PTime(OP,EQ) Mixer1 Mixer2 Reactor1 Packing1 Packing2 Packing3

    ProdA 60 70 120 30 30ProdB 110 240 45 45 60ProdC 80 80 150 40 40 40

  • RTN discrete-time model (GAMS) Few lines of code

    generate: Resources

    Time availability parameters Subsets for domain

    constraints

    Tasks Structural parameters

    Processing times Timing of resource

    consumption/production

    Actual model quite simple

    March 9, 2016 6Enterprise-Wide Optimization Meeting: ABB Project Overview

  • Computational studies How far can we go in terms of problem size?

    Vary # batches per product (1:1:1 A,B,C) Short-term scheduling models, minimizing makespan

    Discrete-time (DT), 5 min slots More batches multiple executions of the same task

    Continuous-time with multiple time grids (CT) More batches more variables and constraints (batch index)

    Discrete-time much better approach But we want 100+ batches, sometimes 500!

    March 9, 2016 7Enterprise-Wide Optimization Meeting: ABB Project Overview

    # batches Makespan (min) DT CPUs CT CPUs1 610 3.36 0.412 1110 1.52 2.313 1620 2.34 65.15 2640 12.8 >1600 (gap=71.3%)10 5190 33.2 -20 10290 397 -

  • How about relying on periodic scheduling? Step 1: Find optimal cyclic pattern

    Given batch proportion between different recipes (ex. 1:1:1)

    Step 2: Solve short-term schedulingfor a given number of repetitions

    Find optimal start-up and shutdown phases Almost all binary variables are fixed Timing of events still free to vary

    March 9, 2016 8Enterprise-Wide Optimization Meeting: ABB Project Overview

    M

    R

    P

    M

    R

    P

    M

    R

    P

    Cycle time= 510 min

    M

    R

    P

    M

    R

    P

    M

    R

    P

    M

    R

    P

    M

    R

    P

    M

    R

    P

    Repeating pattern

    M

    R

    P

    M

    R

    P

    M

    R

    P

    M M

    R R R

    M

    P P P

    Start-up Shut-down

  • Results for periodic-based scheduling Advantages

    Exact same solution as short-termdiscrete-time model

    Continuous-time is more accurate Better computational performance

    Disadvantages Periodic scheduling is for

    stable production of a few products Underlying model works just for a

    specific type of plant topology

    Why not periodic-based schedulingwith discrete-time? Need narrow time window for task

    execution Difficult to generate because start-up

    phase determined by optimization

    March 9, 2016 9Enterprise-Wide Optimization Meeting: ABB Project Overview

    # batches Makespan (min) Total CPUs5 2640 2.210 5190 2.820 10290 3.630 15390 4.350 25590 7.9100 51090 36.2

    # batches Binary variablesTotal

    variables Equations

    5 53 751 110710 53 2851 401720 53 11101 1523730 53 24751 3365750 53 68251 92097100 53 271501 364197

  • Conclusions Guidelines for generating RTN-based

    scheduling model from ISA-95 data

    Existing generic models simply not good enoughto allow tackling problems of industrial relevance Need for optimization based decomposition strategy

    that is competitive with simple heuristics

    Concept tested by repeating pattern from periodic scheduling Continuous-time approach allows for more freedom

    Using just a few binary variables

    Future work Develop a generic and efficient matheuristic

    Integration of (meta)heuristics and mathematical programming Test in a set of real-life problems

    March 9, 2016 10Enterprise-Wide Optimization Meeting: ABB Project Overview

    ISA-95 Friendly RTN Models for Industrial Production SchedulingMotivationGeneric scheduling approachISA-95 data implementationMapping data with RTN process modelRTN discrete-time model (GAMS)Computational studiesHow about relying on periodic scheduling?Results for periodic-based schedulingConclusions

Recommended

View more >